Leveraging ChatGPT for Efficient Fact-Checking
Emma Hoes, Sacha Altay, J. P. Bermeo
Abstract
We explore the potential of automated online content moderation through Large Language Models (LLMs) as a remedy to the overwhelming surge of online content that fact-checking organizations cannot verify. Although LLMs, such as ChatGPT, may exacerbate this problem by facilitating content production, they could also be leveraged to enhance the efficiency and expediency of fact-checking processes. We conducted a systematic evaluation to measure ChatGPT’s fact-checking performance by submitting 21,152 fact-checked statements to ChatGPT as a zero-shot classification. We find that ChatGPT is able to accurately categorize statements as true or false in 69% of cases. Addressing memorization, the performance of ChatGPT is similar on claims that have not been fact-checked or are post its training data cutoff date. These findings demonstrate the potential of ChatGPT to help label misinformation and deepen our comprehension of how LLMs could improve content moderation practices, complementing the crucial work of human fact-checking experts in upholding the accuracy of information dissemination.